Narration of Light: Narration of Light: Computational Tools for Framing the Tonal Imagination [2006]  In my MIT Masters thesis entitled Narration of Light: Computational Tools for Framing the Tonal Imagination, my primary goal was to make computers produce synthetically created volumetric images (commonly referred to as renderings) in accordance with users’ aesthetic preferences. The aim was to develop a system that was not only able to see what the user saw but that also guess the images that would be liked by the user. In other words, I tried to grant eyes to a rendering engine to pick and choose visually well-composed and well-lit images for me.

I developed two systems, one of which was set to evaluate the compositional qualities of a given image, and the other to evaluate the tonal distributions (lighter and darker areas) across an image.

1..            The designer selects images, photographs or renders of interior spaces of his or her liking [in terms of tonal distribution]. The user has to be aware of his or her selection, and the image selected should represent the tones that are admired by the user, or desired for the expected renders. [Random image sections] will not inform the program about the light qualities in the space.
2.            The designer creates a sketchy digital model of the space and imports it to 3DS Max Environment.
The process:
1.            Reading the source image in [terms] of tonal distribution. This helps define various thresholds for [determining the tonal characteristics of] rendered images.
2.            Rendering: The plug-in renders the user defined sequence of images to check [both the compositional and tonal quality] of the rendered frames. This is the initial rendering in which smaller images are rapidly created. All the values are sorted in arrays. [And better images are selected].
3.            Once the sequence is over, the values are checked and the selected images are rendered in high resolution and saved.

Full MIT SMArchS Thesis can be found HERE.


UI Developed in 3DS Max:


Vermeer's paintings, grey level histogram comparisons:

Grey level histogram implemented in 3DS max. Lighter and darker pixel groups can be written as new bitmap files:


Isolated grey level groups are tested with the rule of thirds mask:


Various edge/area detection results. Changing the treshold in the edge detection algorithm (Nevatia-Babu compass) helps cover different spans.